Best Practices

beginner

Six Pillars of Good Genomics Practice

🔄ReproducibilityGit version control; pinned software versions; Nextflow -resume; config files not hardcoded paths

✅Data QualityFastQC before processing; VerifyBamID2 contamination; coverage checks; Ts/Tv monitoring

📋Variant ReportingSanger confirmation; HGVS nomenclature; ACMG criteria documented; VUS annual review

🔐Data SecurityEncrypted storage; HIPAA/GDPR compliance; de-identification; data access agreements

📚DocumentationJupyter/RMarkdown notebooks; parameter logs; data provenance tracking; lab notebook

👥Code ReviewPeer review all pipeline changes; clinical consequences of bugs; test with known samples

Version control everything: Git commit all scripts, configs, and analysis notebooks - not just code
Pin software versions: conda env export > environment.yml; docker build with specific tags (bwa:0.7.17)
Workflow management: Nextflow/Snakemake with checksums on inputs; -resume/--rerun-incomplete for failed jobs
Document parameters: save all CLI flags used; use config files, not hardcoded paths
Lab notebook: Jupyter/RMarkdown for analysis with narrative, code, and figures together
Data provenance: track which pipeline version processed which samples; store in LIMS or database

Raw QC: always run FastQC before ANY processing; fail samples with Q30 <80% or unusual GC distribution
Coverage checks: verify mean depth AND uniformity before calling variants; don't assume coverage from file size
Contamination: run VerifyBamID2 on every clinical sample; FREEMIX >0.03 = investigate/fail
Ts/Tv monitoring: SNP Ts/Tv <2.5 for WES = likely quality issue; check filter settings
Sex concordance: infer genetic sex from X heterozygosity; mismatch = sample swap
Relatedness checks: in multi-sample studies, calculate kinship coefficients (king, PLINK --genome)

Always confirm reportable variants by Sanger sequencing or orthogonal method
Never report a variant without applying all relevant ACMG criteria
Use HGVS nomenclature correctly: NM_007294.4(BRCA1):c.5266dupC p.(Gln1756Profs*74)
Disclose limitations: variants in pseudogenes, repeat regions, or low coverage may be missed
Document evidence: save literature references and database screenshots for each classified variant
Re-interpretation: review VUS variants annually as new evidence emerges; proactive recontact policy
Never phone a pathogenic variant result - always written report reviewed by clinical geneticist

Never store patient genomic data on personal devices or public cloud without encryption + consent
GDPR/HIPAA compliance: de-identify data before sharing; use data access agreements for public datasets
Backup: 3-2-1 rule - 3 copies, 2 different media, 1 offsite; test restores quarterly
Resource management: don't run GATK with default -Xmx4g on a 50 GB genome; profile memory usage
Cluster etiquette: request appropriate resources; kill stuck jobs; don't monopolise shared nodes
Code review: bioinformatics bugs have clinical consequences - peer-review all pipeline changes